Loading...
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/mempool.h>
16#include <linux/gfs2_ondisk.h>
17#include <linux/bio.h>
18#include <linux/fs.h>
19#include <linux/list_sort.h>
20
21#include "dir.h"
22#include "gfs2.h"
23#include "incore.h"
24#include "inode.h"
25#include "glock.h"
26#include "log.h"
27#include "lops.h"
28#include "meta_io.h"
29#include "recovery.h"
30#include "rgrp.h"
31#include "trans.h"
32#include "util.h"
33#include "trace_gfs2.h"
34
35/**
36 * gfs2_pin - Pin a buffer in memory
37 * @sdp: The superblock
38 * @bh: The buffer to be pinned
39 *
40 * The log lock must be held when calling this function
41 */
42void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
43{
44 struct gfs2_bufdata *bd;
45
46 BUG_ON(!current->journal_info);
47
48 clear_buffer_dirty(bh);
49 if (test_set_buffer_pinned(bh))
50 gfs2_assert_withdraw(sdp, 0);
51 if (!buffer_uptodate(bh))
52 gfs2_io_error_bh(sdp, bh);
53 bd = bh->b_private;
54 /* If this buffer is in the AIL and it has already been written
55 * to in-place disk block, remove it from the AIL.
56 */
57 spin_lock(&sdp->sd_ail_lock);
58 if (bd->bd_tr)
59 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
60 spin_unlock(&sdp->sd_ail_lock);
61 get_bh(bh);
62 atomic_inc(&sdp->sd_log_pinned);
63 trace_gfs2_pin(bd, 1);
64}
65
66static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
67{
68 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
69}
70
71static void maybe_release_space(struct gfs2_bufdata *bd)
72{
73 struct gfs2_glock *gl = bd->bd_gl;
74 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
75 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
76 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
77 struct gfs2_bitmap *bi = rgd->rd_bits + index;
78
79 if (bi->bi_clone == NULL)
80 return;
81 if (sdp->sd_args.ar_discard)
82 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
83 memcpy(bi->bi_clone + bi->bi_offset,
84 bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
85 clear_bit(GBF_FULL, &bi->bi_flags);
86 rgd->rd_free_clone = rgd->rd_free;
87 rgd->rd_extfail_pt = rgd->rd_free;
88}
89
90/**
91 * gfs2_unpin - Unpin a buffer
92 * @sdp: the filesystem the buffer belongs to
93 * @bh: The buffer to unpin
94 * @ai:
95 * @flags: The inode dirty flags
96 *
97 */
98
99static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
100 struct gfs2_trans *tr)
101{
102 struct gfs2_bufdata *bd = bh->b_private;
103
104 BUG_ON(!buffer_uptodate(bh));
105 BUG_ON(!buffer_pinned(bh));
106
107 lock_buffer(bh);
108 mark_buffer_dirty(bh);
109 clear_buffer_pinned(bh);
110
111 if (buffer_is_rgrp(bd))
112 maybe_release_space(bd);
113
114 spin_lock(&sdp->sd_ail_lock);
115 if (bd->bd_tr) {
116 list_del(&bd->bd_ail_st_list);
117 brelse(bh);
118 } else {
119 struct gfs2_glock *gl = bd->bd_gl;
120 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
121 atomic_inc(&gl->gl_ail_count);
122 }
123 bd->bd_tr = tr;
124 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
125 spin_unlock(&sdp->sd_ail_lock);
126
127 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
128 trace_gfs2_pin(bd, 0);
129 unlock_buffer(bh);
130 atomic_dec(&sdp->sd_log_pinned);
131}
132
133static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
134{
135 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
136 (sdp->sd_log_flush_head != sdp->sd_log_head));
137
138 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
139 sdp->sd_log_flush_head = 0;
140}
141
142u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
143{
144 unsigned int lbn = sdp->sd_log_flush_head;
145 struct gfs2_journal_extent *je;
146 u64 block;
147
148 list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
149 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
150 block = je->dblock + lbn - je->lblock;
151 gfs2_log_incr_head(sdp);
152 return block;
153 }
154 }
155
156 return -1;
157}
158
159/**
160 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
161 * @sdp: The superblock
162 * @bvec: The bio_vec
163 * @error: The i/o status
164 *
165 * This finds the relevant buffers and unlocks them and sets the
166 * error flag according to the status of the i/o request. This is
167 * used when the log is writing data which has an in-place version
168 * that is pinned in the pagecache.
169 */
170
171static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
172 blk_status_t error)
173{
174 struct buffer_head *bh, *next;
175 struct page *page = bvec->bv_page;
176 unsigned size;
177
178 bh = page_buffers(page);
179 size = bvec->bv_len;
180 while (bh_offset(bh) < bvec->bv_offset)
181 bh = bh->b_this_page;
182 do {
183 if (error)
184 mark_buffer_write_io_error(bh);
185 unlock_buffer(bh);
186 next = bh->b_this_page;
187 size -= bh->b_size;
188 brelse(bh);
189 bh = next;
190 } while(bh && size);
191}
192
193/**
194 * gfs2_end_log_write - end of i/o to the log
195 * @bio: The bio
196 * @error: Status of i/o request
197 *
198 * Each bio_vec contains either data from the pagecache or data
199 * relating to the log itself. Here we iterate over the bio_vec
200 * array, processing both kinds of data.
201 *
202 */
203
204static void gfs2_end_log_write(struct bio *bio)
205{
206 struct gfs2_sbd *sdp = bio->bi_private;
207 struct bio_vec *bvec;
208 struct page *page;
209 int i;
210
211 if (bio->bi_status) {
212 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
213 bio->bi_status, sdp->sd_jdesc->jd_jid);
214 wake_up(&sdp->sd_logd_waitq);
215 }
216
217 bio_for_each_segment_all(bvec, bio, i) {
218 page = bvec->bv_page;
219 if (page_has_buffers(page))
220 gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
221 else
222 mempool_free(page, gfs2_page_pool);
223 }
224
225 bio_put(bio);
226 if (atomic_dec_and_test(&sdp->sd_log_in_flight))
227 wake_up(&sdp->sd_log_flush_wait);
228}
229
230/**
231 * gfs2_log_flush_bio - Submit any pending log bio
232 * @sdp: The superblock
233 * @op: REQ_OP
234 * @op_flags: req_flag_bits
235 *
236 * Submit any pending part-built or full bio to the block device. If
237 * there is no pending bio, then this is a no-op.
238 */
239
240void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int op, int op_flags)
241{
242 if (sdp->sd_log_bio) {
243 atomic_inc(&sdp->sd_log_in_flight);
244 bio_set_op_attrs(sdp->sd_log_bio, op, op_flags);
245 submit_bio(sdp->sd_log_bio);
246 sdp->sd_log_bio = NULL;
247 }
248}
249
250/**
251 * gfs2_log_alloc_bio - Allocate a new bio for log writing
252 * @sdp: The superblock
253 * @blkno: The next device block number we want to write to
254 *
255 * This should never be called when there is a cached bio in the
256 * super block. When it returns, there will be a cached bio in the
257 * super block which will have as many bio_vecs as the device is
258 * happy to handle.
259 *
260 * Returns: Newly allocated bio
261 */
262
263static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
264{
265 struct super_block *sb = sdp->sd_vfs;
266 struct bio *bio;
267
268 BUG_ON(sdp->sd_log_bio);
269
270 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
271 bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
272 bio_set_dev(bio, sb->s_bdev);
273 bio->bi_end_io = gfs2_end_log_write;
274 bio->bi_private = sdp;
275
276 sdp->sd_log_bio = bio;
277
278 return bio;
279}
280
281/**
282 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
283 * @sdp: The superblock
284 * @blkno: The device block number we want to write to
285 *
286 * If there is a cached bio, then if the next block number is sequential
287 * with the previous one, return it, otherwise flush the bio to the
288 * device. If there is not a cached bio, or we just flushed it, then
289 * allocate a new one.
290 *
291 * Returns: The bio to use for log writes
292 */
293
294static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
295{
296 struct bio *bio = sdp->sd_log_bio;
297 u64 nblk;
298
299 if (bio) {
300 nblk = bio_end_sector(bio);
301 nblk >>= sdp->sd_fsb2bb_shift;
302 if (blkno == nblk)
303 return bio;
304 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
305 }
306
307 return gfs2_log_alloc_bio(sdp, blkno);
308}
309
310/**
311 * gfs2_log_write - write to log
312 * @sdp: the filesystem
313 * @page: the page to write
314 * @size: the size of the data to write
315 * @offset: the offset within the page
316 * @blkno: block number of the log entry
317 *
318 * Try and add the page segment to the current bio. If that fails,
319 * submit the current bio to the device and create a new one, and
320 * then add the page segment to that.
321 */
322
323void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
324 unsigned size, unsigned offset, u64 blkno)
325{
326 struct bio *bio;
327 int ret;
328
329 bio = gfs2_log_get_bio(sdp, blkno);
330 ret = bio_add_page(bio, page, size, offset);
331 if (ret == 0) {
332 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
333 bio = gfs2_log_alloc_bio(sdp, blkno);
334 ret = bio_add_page(bio, page, size, offset);
335 WARN_ON(ret == 0);
336 }
337}
338
339/**
340 * gfs2_log_write_bh - write a buffer's content to the log
341 * @sdp: The super block
342 * @bh: The buffer pointing to the in-place location
343 *
344 * This writes the content of the buffer to the next available location
345 * in the log. The buffer will be unlocked once the i/o to the log has
346 * completed.
347 */
348
349static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
350{
351 gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh),
352 gfs2_log_bmap(sdp));
353}
354
355/**
356 * gfs2_log_write_page - write one block stored in a page, into the log
357 * @sdp: The superblock
358 * @page: The struct page
359 *
360 * This writes the first block-sized part of the page into the log. Note
361 * that the page must have been allocated from the gfs2_page_pool mempool
362 * and that after this has been called, ownership has been transferred and
363 * the page may be freed at any time.
364 */
365
366void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
367{
368 struct super_block *sb = sdp->sd_vfs;
369 gfs2_log_write(sdp, page, sb->s_blocksize, 0,
370 gfs2_log_bmap(sdp));
371}
372
373static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
374 u32 ld_length, u32 ld_data1)
375{
376 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
377 struct gfs2_log_descriptor *ld = page_address(page);
378 clear_page(ld);
379 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
380 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
381 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
382 ld->ld_type = cpu_to_be32(ld_type);
383 ld->ld_length = cpu_to_be32(ld_length);
384 ld->ld_data1 = cpu_to_be32(ld_data1);
385 ld->ld_data2 = 0;
386 return page;
387}
388
389static void gfs2_check_magic(struct buffer_head *bh)
390{
391 void *kaddr;
392 __be32 *ptr;
393
394 clear_buffer_escaped(bh);
395 kaddr = kmap_atomic(bh->b_page);
396 ptr = kaddr + bh_offset(bh);
397 if (*ptr == cpu_to_be32(GFS2_MAGIC))
398 set_buffer_escaped(bh);
399 kunmap_atomic(kaddr);
400}
401
402static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
403{
404 struct gfs2_bufdata *bda, *bdb;
405
406 bda = list_entry(a, struct gfs2_bufdata, bd_list);
407 bdb = list_entry(b, struct gfs2_bufdata, bd_list);
408
409 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
410 return -1;
411 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
412 return 1;
413 return 0;
414}
415
416static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
417 unsigned int total, struct list_head *blist,
418 bool is_databuf)
419{
420 struct gfs2_log_descriptor *ld;
421 struct gfs2_bufdata *bd1 = NULL, *bd2;
422 struct page *page;
423 unsigned int num;
424 unsigned n;
425 __be64 *ptr;
426
427 gfs2_log_lock(sdp);
428 list_sort(NULL, blist, blocknr_cmp);
429 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
430 while(total) {
431 num = total;
432 if (total > limit)
433 num = limit;
434 gfs2_log_unlock(sdp);
435 page = gfs2_get_log_desc(sdp,
436 is_databuf ? GFS2_LOG_DESC_JDATA :
437 GFS2_LOG_DESC_METADATA, num + 1, num);
438 ld = page_address(page);
439 gfs2_log_lock(sdp);
440 ptr = (__be64 *)(ld + 1);
441
442 n = 0;
443 list_for_each_entry_continue(bd1, blist, bd_list) {
444 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
445 if (is_databuf) {
446 gfs2_check_magic(bd1->bd_bh);
447 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
448 }
449 if (++n >= num)
450 break;
451 }
452
453 gfs2_log_unlock(sdp);
454 gfs2_log_write_page(sdp, page);
455 gfs2_log_lock(sdp);
456
457 n = 0;
458 list_for_each_entry_continue(bd2, blist, bd_list) {
459 get_bh(bd2->bd_bh);
460 gfs2_log_unlock(sdp);
461 lock_buffer(bd2->bd_bh);
462
463 if (buffer_escaped(bd2->bd_bh)) {
464 void *kaddr;
465 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
466 ptr = page_address(page);
467 kaddr = kmap_atomic(bd2->bd_bh->b_page);
468 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
469 bd2->bd_bh->b_size);
470 kunmap_atomic(kaddr);
471 *(__be32 *)ptr = 0;
472 clear_buffer_escaped(bd2->bd_bh);
473 unlock_buffer(bd2->bd_bh);
474 brelse(bd2->bd_bh);
475 gfs2_log_write_page(sdp, page);
476 } else {
477 gfs2_log_write_bh(sdp, bd2->bd_bh);
478 }
479 gfs2_log_lock(sdp);
480 if (++n >= num)
481 break;
482 }
483
484 BUG_ON(total < num);
485 total -= num;
486 }
487 gfs2_log_unlock(sdp);
488}
489
490static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
491{
492 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
493 unsigned int nbuf;
494 if (tr == NULL)
495 return;
496 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
497 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
498}
499
500static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
501{
502 struct list_head *head;
503 struct gfs2_bufdata *bd;
504
505 if (tr == NULL)
506 return;
507
508 head = &tr->tr_buf;
509 while (!list_empty(head)) {
510 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
511 list_del_init(&bd->bd_list);
512 gfs2_unpin(sdp, bd->bd_bh, tr);
513 }
514}
515
516static void buf_lo_before_scan(struct gfs2_jdesc *jd,
517 struct gfs2_log_header_host *head, int pass)
518{
519 if (pass != 0)
520 return;
521
522 jd->jd_found_blocks = 0;
523 jd->jd_replayed_blocks = 0;
524}
525
526static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
527 struct gfs2_log_descriptor *ld, __be64 *ptr,
528 int pass)
529{
530 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
531 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
532 struct gfs2_glock *gl = ip->i_gl;
533 unsigned int blks = be32_to_cpu(ld->ld_data1);
534 struct buffer_head *bh_log, *bh_ip;
535 u64 blkno;
536 int error = 0;
537
538 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
539 return 0;
540
541 gfs2_replay_incr_blk(jd, &start);
542
543 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
544 blkno = be64_to_cpu(*ptr++);
545
546 jd->jd_found_blocks++;
547
548 if (gfs2_revoke_check(jd, blkno, start))
549 continue;
550
551 error = gfs2_replay_read_block(jd, start, &bh_log);
552 if (error)
553 return error;
554
555 bh_ip = gfs2_meta_new(gl, blkno);
556 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
557
558 if (gfs2_meta_check(sdp, bh_ip))
559 error = -EIO;
560 else
561 mark_buffer_dirty(bh_ip);
562
563 brelse(bh_log);
564 brelse(bh_ip);
565
566 if (error)
567 break;
568
569 jd->jd_replayed_blocks++;
570 }
571
572 return error;
573}
574
575/**
576 * gfs2_meta_sync - Sync all buffers associated with a glock
577 * @gl: The glock
578 *
579 */
580
581static void gfs2_meta_sync(struct gfs2_glock *gl)
582{
583 struct address_space *mapping = gfs2_glock2aspace(gl);
584 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
585 int error;
586
587 if (mapping == NULL)
588 mapping = &sdp->sd_aspace;
589
590 filemap_fdatawrite(mapping);
591 error = filemap_fdatawait(mapping);
592
593 if (error)
594 gfs2_io_error(gl->gl_name.ln_sbd);
595}
596
597static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
598{
599 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
600 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
601
602 if (error) {
603 gfs2_meta_sync(ip->i_gl);
604 return;
605 }
606 if (pass != 1)
607 return;
608
609 gfs2_meta_sync(ip->i_gl);
610
611 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
612 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
613}
614
615static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
616{
617 struct gfs2_meta_header *mh;
618 unsigned int offset;
619 struct list_head *head = &sdp->sd_log_le_revoke;
620 struct gfs2_bufdata *bd;
621 struct page *page;
622 unsigned int length;
623
624 gfs2_write_revokes(sdp);
625 if (!sdp->sd_log_num_revoke)
626 return;
627
628 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
629 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
630 offset = sizeof(struct gfs2_log_descriptor);
631
632 list_for_each_entry(bd, head, bd_list) {
633 sdp->sd_log_num_revoke--;
634
635 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
636
637 gfs2_log_write_page(sdp, page);
638 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
639 mh = page_address(page);
640 clear_page(mh);
641 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
642 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
643 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
644 offset = sizeof(struct gfs2_meta_header);
645 }
646
647 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
648 offset += sizeof(u64);
649 }
650 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
651
652 gfs2_log_write_page(sdp, page);
653}
654
655static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
656{
657 struct list_head *head = &sdp->sd_log_le_revoke;
658 struct gfs2_bufdata *bd;
659 struct gfs2_glock *gl;
660
661 while (!list_empty(head)) {
662 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
663 list_del_init(&bd->bd_list);
664 gl = bd->bd_gl;
665 atomic_dec(&gl->gl_revokes);
666 clear_bit(GLF_LFLUSH, &gl->gl_flags);
667 kmem_cache_free(gfs2_bufdata_cachep, bd);
668 }
669}
670
671static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
672 struct gfs2_log_header_host *head, int pass)
673{
674 if (pass != 0)
675 return;
676
677 jd->jd_found_revokes = 0;
678 jd->jd_replay_tail = head->lh_tail;
679}
680
681static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
682 struct gfs2_log_descriptor *ld, __be64 *ptr,
683 int pass)
684{
685 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
686 unsigned int blks = be32_to_cpu(ld->ld_length);
687 unsigned int revokes = be32_to_cpu(ld->ld_data1);
688 struct buffer_head *bh;
689 unsigned int offset;
690 u64 blkno;
691 int first = 1;
692 int error;
693
694 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
695 return 0;
696
697 offset = sizeof(struct gfs2_log_descriptor);
698
699 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
700 error = gfs2_replay_read_block(jd, start, &bh);
701 if (error)
702 return error;
703
704 if (!first)
705 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
706
707 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
708 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
709
710 error = gfs2_revoke_add(jd, blkno, start);
711 if (error < 0) {
712 brelse(bh);
713 return error;
714 }
715 else if (error)
716 jd->jd_found_revokes++;
717
718 if (!--revokes)
719 break;
720 offset += sizeof(u64);
721 }
722
723 brelse(bh);
724 offset = sizeof(struct gfs2_meta_header);
725 first = 0;
726 }
727
728 return 0;
729}
730
731static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
732{
733 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
734
735 if (error) {
736 gfs2_revoke_clean(jd);
737 return;
738 }
739 if (pass != 1)
740 return;
741
742 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
743 jd->jd_jid, jd->jd_found_revokes);
744
745 gfs2_revoke_clean(jd);
746}
747
748/**
749 * databuf_lo_before_commit - Scan the data buffers, writing as we go
750 *
751 */
752
753static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
754{
755 unsigned int limit = databuf_limit(sdp);
756 unsigned int nbuf;
757 if (tr == NULL)
758 return;
759 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
760 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
761}
762
763static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
764 struct gfs2_log_descriptor *ld,
765 __be64 *ptr, int pass)
766{
767 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
768 struct gfs2_glock *gl = ip->i_gl;
769 unsigned int blks = be32_to_cpu(ld->ld_data1);
770 struct buffer_head *bh_log, *bh_ip;
771 u64 blkno;
772 u64 esc;
773 int error = 0;
774
775 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
776 return 0;
777
778 gfs2_replay_incr_blk(jd, &start);
779 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
780 blkno = be64_to_cpu(*ptr++);
781 esc = be64_to_cpu(*ptr++);
782
783 jd->jd_found_blocks++;
784
785 if (gfs2_revoke_check(jd, blkno, start))
786 continue;
787
788 error = gfs2_replay_read_block(jd, start, &bh_log);
789 if (error)
790 return error;
791
792 bh_ip = gfs2_meta_new(gl, blkno);
793 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
794
795 /* Unescape */
796 if (esc) {
797 __be32 *eptr = (__be32 *)bh_ip->b_data;
798 *eptr = cpu_to_be32(GFS2_MAGIC);
799 }
800 mark_buffer_dirty(bh_ip);
801
802 brelse(bh_log);
803 brelse(bh_ip);
804
805 jd->jd_replayed_blocks++;
806 }
807
808 return error;
809}
810
811/* FIXME: sort out accounting for log blocks etc. */
812
813static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
814{
815 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
816 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
817
818 if (error) {
819 gfs2_meta_sync(ip->i_gl);
820 return;
821 }
822 if (pass != 1)
823 return;
824
825 /* data sync? */
826 gfs2_meta_sync(ip->i_gl);
827
828 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
829 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
830}
831
832static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
833{
834 struct list_head *head;
835 struct gfs2_bufdata *bd;
836
837 if (tr == NULL)
838 return;
839
840 head = &tr->tr_databuf;
841 while (!list_empty(head)) {
842 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
843 list_del_init(&bd->bd_list);
844 gfs2_unpin(sdp, bd->bd_bh, tr);
845 }
846}
847
848
849const struct gfs2_log_operations gfs2_buf_lops = {
850 .lo_before_commit = buf_lo_before_commit,
851 .lo_after_commit = buf_lo_after_commit,
852 .lo_before_scan = buf_lo_before_scan,
853 .lo_scan_elements = buf_lo_scan_elements,
854 .lo_after_scan = buf_lo_after_scan,
855 .lo_name = "buf",
856};
857
858const struct gfs2_log_operations gfs2_revoke_lops = {
859 .lo_before_commit = revoke_lo_before_commit,
860 .lo_after_commit = revoke_lo_after_commit,
861 .lo_before_scan = revoke_lo_before_scan,
862 .lo_scan_elements = revoke_lo_scan_elements,
863 .lo_after_scan = revoke_lo_after_scan,
864 .lo_name = "revoke",
865};
866
867const struct gfs2_log_operations gfs2_databuf_lops = {
868 .lo_before_commit = databuf_lo_before_commit,
869 .lo_after_commit = databuf_lo_after_commit,
870 .lo_scan_elements = databuf_lo_scan_elements,
871 .lo_after_scan = databuf_lo_after_scan,
872 .lo_name = "databuf",
873};
874
875const struct gfs2_log_operations *gfs2_log_ops[] = {
876 &gfs2_databuf_lops,
877 &gfs2_buf_lops,
878 &gfs2_revoke_lops,
879 NULL,
880};
881
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 */
6
7#include <linux/sched.h>
8#include <linux/slab.h>
9#include <linux/spinlock.h>
10#include <linux/completion.h>
11#include <linux/buffer_head.h>
12#include <linux/mempool.h>
13#include <linux/gfs2_ondisk.h>
14#include <linux/bio.h>
15#include <linux/fs.h>
16#include <linux/list_sort.h>
17#include <linux/blkdev.h>
18
19#include "bmap.h"
20#include "dir.h"
21#include "gfs2.h"
22#include "incore.h"
23#include "inode.h"
24#include "glock.h"
25#include "glops.h"
26#include "log.h"
27#include "lops.h"
28#include "meta_io.h"
29#include "recovery.h"
30#include "rgrp.h"
31#include "trans.h"
32#include "util.h"
33#include "trace_gfs2.h"
34
35/**
36 * gfs2_pin - Pin a buffer in memory
37 * @sdp: The superblock
38 * @bh: The buffer to be pinned
39 *
40 * The log lock must be held when calling this function
41 */
42void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
43{
44 struct gfs2_bufdata *bd;
45
46 BUG_ON(!current->journal_info);
47
48 clear_buffer_dirty(bh);
49 if (test_set_buffer_pinned(bh))
50 gfs2_assert_withdraw(sdp, 0);
51 if (!buffer_uptodate(bh))
52 gfs2_io_error_bh_wd(sdp, bh);
53 bd = bh->b_private;
54 /* If this buffer is in the AIL and it has already been written
55 * to in-place disk block, remove it from the AIL.
56 */
57 spin_lock(&sdp->sd_ail_lock);
58 if (bd->bd_tr)
59 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
60 spin_unlock(&sdp->sd_ail_lock);
61 get_bh(bh);
62 atomic_inc(&sdp->sd_log_pinned);
63 trace_gfs2_pin(bd, 1);
64}
65
66static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
67{
68 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
69}
70
71static void maybe_release_space(struct gfs2_bufdata *bd)
72{
73 struct gfs2_glock *gl = bd->bd_gl;
74 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
75 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
76 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
77 struct gfs2_bitmap *bi = rgd->rd_bits + index;
78
79 rgrp_lock_local(rgd);
80 if (bi->bi_clone == NULL)
81 goto out;
82 if (sdp->sd_args.ar_discard)
83 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
84 memcpy(bi->bi_clone + bi->bi_offset,
85 bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
86 clear_bit(GBF_FULL, &bi->bi_flags);
87 rgd->rd_free_clone = rgd->rd_free;
88 BUG_ON(rgd->rd_free_clone < rgd->rd_reserved);
89 rgd->rd_extfail_pt = rgd->rd_free;
90
91out:
92 rgrp_unlock_local(rgd);
93}
94
95/**
96 * gfs2_unpin - Unpin a buffer
97 * @sdp: the filesystem the buffer belongs to
98 * @bh: The buffer to unpin
99 * @tr: The system transaction being flushed
100 */
101
102static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
103 struct gfs2_trans *tr)
104{
105 struct gfs2_bufdata *bd = bh->b_private;
106
107 BUG_ON(!buffer_uptodate(bh));
108 BUG_ON(!buffer_pinned(bh));
109
110 lock_buffer(bh);
111 mark_buffer_dirty(bh);
112 clear_buffer_pinned(bh);
113
114 if (buffer_is_rgrp(bd))
115 maybe_release_space(bd);
116
117 spin_lock(&sdp->sd_ail_lock);
118 if (bd->bd_tr) {
119 list_del(&bd->bd_ail_st_list);
120 brelse(bh);
121 } else {
122 struct gfs2_glock *gl = bd->bd_gl;
123 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
124 atomic_inc(&gl->gl_ail_count);
125 }
126 bd->bd_tr = tr;
127 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
128 spin_unlock(&sdp->sd_ail_lock);
129
130 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
131 trace_gfs2_pin(bd, 0);
132 unlock_buffer(bh);
133 atomic_dec(&sdp->sd_log_pinned);
134}
135
136void gfs2_log_incr_head(struct gfs2_sbd *sdp)
137{
138 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
139 (sdp->sd_log_flush_head != sdp->sd_log_head));
140
141 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
142 sdp->sd_log_flush_head = 0;
143}
144
145u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock)
146{
147 struct gfs2_journal_extent *je;
148
149 list_for_each_entry(je, &jd->extent_list, list) {
150 if (lblock >= je->lblock && lblock < je->lblock + je->blocks)
151 return je->dblock + lblock - je->lblock;
152 }
153
154 return -1;
155}
156
157/**
158 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
159 * @sdp: The superblock
160 * @bvec: The bio_vec
161 * @error: The i/o status
162 *
163 * This finds the relevant buffers and unlocks them and sets the
164 * error flag according to the status of the i/o request. This is
165 * used when the log is writing data which has an in-place version
166 * that is pinned in the pagecache.
167 */
168
169static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
170 struct bio_vec *bvec,
171 blk_status_t error)
172{
173 struct buffer_head *bh, *next;
174 struct page *page = bvec->bv_page;
175 unsigned size;
176
177 bh = page_buffers(page);
178 size = bvec->bv_len;
179 while (bh_offset(bh) < bvec->bv_offset)
180 bh = bh->b_this_page;
181 do {
182 if (error)
183 mark_buffer_write_io_error(bh);
184 unlock_buffer(bh);
185 next = bh->b_this_page;
186 size -= bh->b_size;
187 brelse(bh);
188 bh = next;
189 } while(bh && size);
190}
191
192/**
193 * gfs2_end_log_write - end of i/o to the log
194 * @bio: The bio
195 *
196 * Each bio_vec contains either data from the pagecache or data
197 * relating to the log itself. Here we iterate over the bio_vec
198 * array, processing both kinds of data.
199 *
200 */
201
202static void gfs2_end_log_write(struct bio *bio)
203{
204 struct gfs2_sbd *sdp = bio->bi_private;
205 struct bio_vec *bvec;
206 struct page *page;
207 struct bvec_iter_all iter_all;
208
209 if (bio->bi_status) {
210 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status))
211 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
212 bio->bi_status, sdp->sd_jdesc->jd_jid);
213 gfs2_withdraw_delayed(sdp);
214 /* prevent more writes to the journal */
215 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
216 wake_up(&sdp->sd_logd_waitq);
217 }
218
219 bio_for_each_segment_all(bvec, bio, iter_all) {
220 page = bvec->bv_page;
221 if (page_has_buffers(page))
222 gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
223 else
224 mempool_free(page, gfs2_page_pool);
225 }
226
227 bio_put(bio);
228 if (atomic_dec_and_test(&sdp->sd_log_in_flight))
229 wake_up(&sdp->sd_log_flush_wait);
230}
231
232/**
233 * gfs2_log_submit_bio - Submit any pending log bio
234 * @biop: Address of the bio pointer
235 * @opf: REQ_OP | op_flags
236 *
237 * Submit any pending part-built or full bio to the block device. If
238 * there is no pending bio, then this is a no-op.
239 */
240
241void gfs2_log_submit_bio(struct bio **biop, blk_opf_t opf)
242{
243 struct bio *bio = *biop;
244 if (bio) {
245 struct gfs2_sbd *sdp = bio->bi_private;
246 atomic_inc(&sdp->sd_log_in_flight);
247 bio->bi_opf = opf;
248 submit_bio(bio);
249 *biop = NULL;
250 }
251}
252
253/**
254 * gfs2_log_alloc_bio - Allocate a bio
255 * @sdp: The super block
256 * @blkno: The device block number we want to write to
257 * @end_io: The bi_end_io callback
258 *
259 * Allocate a new bio, initialize it with the given parameters and return it.
260 *
261 * Returns: The newly allocated bio
262 */
263
264static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
265 bio_end_io_t *end_io)
266{
267 struct super_block *sb = sdp->sd_vfs;
268 struct bio *bio = bio_alloc(sb->s_bdev, BIO_MAX_VECS, 0, GFP_NOIO);
269
270 bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
271 bio->bi_end_io = end_io;
272 bio->bi_private = sdp;
273
274 return bio;
275}
276
277/**
278 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
279 * @sdp: The super block
280 * @blkno: The device block number we want to write to
281 * @biop: The bio to get or allocate
282 * @op: REQ_OP
283 * @end_io: The bi_end_io callback
284 * @flush: Always flush the current bio and allocate a new one?
285 *
286 * If there is a cached bio, then if the next block number is sequential
287 * with the previous one, return it, otherwise flush the bio to the
288 * device. If there is no cached bio, or we just flushed it, then
289 * allocate a new one.
290 *
291 * Returns: The bio to use for log writes
292 */
293
294static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
295 struct bio **biop, enum req_op op,
296 bio_end_io_t *end_io, bool flush)
297{
298 struct bio *bio = *biop;
299
300 if (bio) {
301 u64 nblk;
302
303 nblk = bio_end_sector(bio);
304 nblk >>= sdp->sd_fsb2bb_shift;
305 if (blkno == nblk && !flush)
306 return bio;
307 gfs2_log_submit_bio(biop, op);
308 }
309
310 *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
311 return *biop;
312}
313
314/**
315 * gfs2_log_write - write to log
316 * @sdp: the filesystem
317 * @jd: The journal descriptor
318 * @page: the page to write
319 * @size: the size of the data to write
320 * @offset: the offset within the page
321 * @blkno: block number of the log entry
322 *
323 * Try and add the page segment to the current bio. If that fails,
324 * submit the current bio to the device and create a new one, and
325 * then add the page segment to that.
326 */
327
328void gfs2_log_write(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
329 struct page *page, unsigned size, unsigned offset,
330 u64 blkno)
331{
332 struct bio *bio;
333 int ret;
334
335 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio, REQ_OP_WRITE,
336 gfs2_end_log_write, false);
337 ret = bio_add_page(bio, page, size, offset);
338 if (ret == 0) {
339 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio,
340 REQ_OP_WRITE, gfs2_end_log_write, true);
341 ret = bio_add_page(bio, page, size, offset);
342 WARN_ON(ret == 0);
343 }
344}
345
346/**
347 * gfs2_log_write_bh - write a buffer's content to the log
348 * @sdp: The super block
349 * @bh: The buffer pointing to the in-place location
350 *
351 * This writes the content of the buffer to the next available location
352 * in the log. The buffer will be unlocked once the i/o to the log has
353 * completed.
354 */
355
356static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
357{
358 u64 dblock;
359
360 dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
361 gfs2_log_incr_head(sdp);
362 gfs2_log_write(sdp, sdp->sd_jdesc, bh->b_page, bh->b_size,
363 bh_offset(bh), dblock);
364}
365
366/**
367 * gfs2_log_write_page - write one block stored in a page, into the log
368 * @sdp: The superblock
369 * @page: The struct page
370 *
371 * This writes the first block-sized part of the page into the log. Note
372 * that the page must have been allocated from the gfs2_page_pool mempool
373 * and that after this has been called, ownership has been transferred and
374 * the page may be freed at any time.
375 */
376
377static void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
378{
379 struct super_block *sb = sdp->sd_vfs;
380 u64 dblock;
381
382 dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
383 gfs2_log_incr_head(sdp);
384 gfs2_log_write(sdp, sdp->sd_jdesc, page, sb->s_blocksize, 0, dblock);
385}
386
387/**
388 * gfs2_end_log_read - end I/O callback for reads from the log
389 * @bio: The bio
390 *
391 * Simply unlock the pages in the bio. The main thread will wait on them and
392 * process them in order as necessary.
393 */
394
395static void gfs2_end_log_read(struct bio *bio)
396{
397 struct page *page;
398 struct bio_vec *bvec;
399 struct bvec_iter_all iter_all;
400
401 bio_for_each_segment_all(bvec, bio, iter_all) {
402 page = bvec->bv_page;
403 if (bio->bi_status) {
404 int err = blk_status_to_errno(bio->bi_status);
405
406 SetPageError(page);
407 mapping_set_error(page->mapping, err);
408 }
409 unlock_page(page);
410 }
411
412 bio_put(bio);
413}
414
415/**
416 * gfs2_jhead_pg_srch - Look for the journal head in a given page.
417 * @jd: The journal descriptor
418 * @head: The journal head to start from
419 * @page: The page to look in
420 *
421 * Returns: 1 if found, 0 otherwise.
422 */
423
424static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
425 struct gfs2_log_header_host *head,
426 struct page *page)
427{
428 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
429 struct gfs2_log_header_host lh;
430 void *kaddr = kmap_atomic(page);
431 unsigned int offset;
432 bool ret = false;
433
434 for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
435 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
436 if (lh.lh_sequence >= head->lh_sequence)
437 *head = lh;
438 else {
439 ret = true;
440 break;
441 }
442 }
443 }
444 kunmap_atomic(kaddr);
445 return ret;
446}
447
448/**
449 * gfs2_jhead_process_page - Search/cleanup a page
450 * @jd: The journal descriptor
451 * @index: Index of the page to look into
452 * @head: The journal head to start from
453 * @done: If set, perform only cleanup, else search and set if found.
454 *
455 * Find the folio with 'index' in the journal's mapping. Search the folio for
456 * the journal head if requested (cleanup == false). Release refs on the
457 * folio so the page cache can reclaim it. We grabbed a
458 * reference on this folio twice, first when we did a find_or_create_page()
459 * to obtain the folio to add it to the bio and second when we do a
460 * filemap_get_folio() here to get the folio to wait on while I/O on it is being
461 * completed.
462 * This function is also used to free up a folio we might've grabbed but not
463 * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
464 * submitted the I/O, but we already found the jhead so we only need to drop
465 * our references to the folio.
466 */
467
468static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
469 struct gfs2_log_header_host *head,
470 bool *done)
471{
472 struct folio *folio;
473
474 folio = filemap_get_folio(jd->jd_inode->i_mapping, index);
475
476 folio_wait_locked(folio);
477 if (folio_test_error(folio))
478 *done = true;
479
480 if (!*done)
481 *done = gfs2_jhead_pg_srch(jd, head, &folio->page);
482
483 /* filemap_get_folio() and the earlier find_or_create_page() */
484 folio_put_refs(folio, 2);
485}
486
487static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
488{
489 struct bio *new;
490
491 new = bio_alloc(prev->bi_bdev, nr_iovecs, prev->bi_opf, GFP_NOIO);
492 bio_clone_blkg_association(new, prev);
493 new->bi_iter.bi_sector = bio_end_sector(prev);
494 bio_chain(new, prev);
495 submit_bio(prev);
496 return new;
497}
498
499/**
500 * gfs2_find_jhead - find the head of a log
501 * @jd: The journal descriptor
502 * @head: The log descriptor for the head of the log is returned here
503 * @keep_cache: If set inode pages will not be truncated
504 *
505 * Do a search of a journal by reading it in large chunks using bios and find
506 * the valid log entry with the highest sequence number. (i.e. the log head)
507 *
508 * Returns: 0 on success, errno otherwise
509 */
510int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
511 bool keep_cache)
512{
513 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
514 struct address_space *mapping = jd->jd_inode->i_mapping;
515 unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
516 unsigned int bsize = sdp->sd_sb.sb_bsize, off;
517 unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
518 unsigned int shift = PAGE_SHIFT - bsize_shift;
519 unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift;
520 struct gfs2_journal_extent *je;
521 int sz, ret = 0;
522 struct bio *bio = NULL;
523 struct page *page = NULL;
524 bool done = false;
525 errseq_t since;
526
527 memset(head, 0, sizeof(*head));
528 if (list_empty(&jd->extent_list))
529 gfs2_map_journal_extents(sdp, jd);
530
531 since = filemap_sample_wb_err(mapping);
532 list_for_each_entry(je, &jd->extent_list, list) {
533 u64 dblock = je->dblock;
534
535 for (; block < je->lblock + je->blocks; block++, dblock++) {
536 if (!page) {
537 page = find_or_create_page(mapping,
538 block >> shift, GFP_NOFS);
539 if (!page) {
540 ret = -ENOMEM;
541 done = true;
542 goto out;
543 }
544 off = 0;
545 }
546
547 if (bio && (off || block < blocks_submitted + max_blocks)) {
548 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
549
550 if (bio_end_sector(bio) == sector) {
551 sz = bio_add_page(bio, page, bsize, off);
552 if (sz == bsize)
553 goto block_added;
554 }
555 if (off) {
556 unsigned int blocks =
557 (PAGE_SIZE - off) >> bsize_shift;
558
559 bio = gfs2_chain_bio(bio, blocks);
560 goto add_block_to_new_bio;
561 }
562 }
563
564 if (bio) {
565 blocks_submitted = block;
566 submit_bio(bio);
567 }
568
569 bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
570 bio->bi_opf = REQ_OP_READ;
571add_block_to_new_bio:
572 sz = bio_add_page(bio, page, bsize, off);
573 BUG_ON(sz != bsize);
574block_added:
575 off += bsize;
576 if (off == PAGE_SIZE)
577 page = NULL;
578 if (blocks_submitted <= blocks_read + max_blocks) {
579 /* Keep at least one bio in flight */
580 continue;
581 }
582
583 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
584 blocks_read += PAGE_SIZE >> bsize_shift;
585 if (done)
586 goto out; /* found */
587 }
588 }
589
590out:
591 if (bio)
592 submit_bio(bio);
593 while (blocks_read < block) {
594 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
595 blocks_read += PAGE_SIZE >> bsize_shift;
596 }
597
598 if (!ret)
599 ret = filemap_check_wb_err(mapping, since);
600
601 if (!keep_cache)
602 truncate_inode_pages(mapping, 0);
603
604 return ret;
605}
606
607static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
608 u32 ld_length, u32 ld_data1)
609{
610 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
611 struct gfs2_log_descriptor *ld = page_address(page);
612 clear_page(ld);
613 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
614 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
615 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
616 ld->ld_type = cpu_to_be32(ld_type);
617 ld->ld_length = cpu_to_be32(ld_length);
618 ld->ld_data1 = cpu_to_be32(ld_data1);
619 ld->ld_data2 = 0;
620 return page;
621}
622
623static void gfs2_check_magic(struct buffer_head *bh)
624{
625 void *kaddr;
626 __be32 *ptr;
627
628 clear_buffer_escaped(bh);
629 kaddr = kmap_atomic(bh->b_page);
630 ptr = kaddr + bh_offset(bh);
631 if (*ptr == cpu_to_be32(GFS2_MAGIC))
632 set_buffer_escaped(bh);
633 kunmap_atomic(kaddr);
634}
635
636static int blocknr_cmp(void *priv, const struct list_head *a,
637 const struct list_head *b)
638{
639 struct gfs2_bufdata *bda, *bdb;
640
641 bda = list_entry(a, struct gfs2_bufdata, bd_list);
642 bdb = list_entry(b, struct gfs2_bufdata, bd_list);
643
644 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
645 return -1;
646 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
647 return 1;
648 return 0;
649}
650
651static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
652 unsigned int total, struct list_head *blist,
653 bool is_databuf)
654{
655 struct gfs2_log_descriptor *ld;
656 struct gfs2_bufdata *bd1 = NULL, *bd2;
657 struct page *page;
658 unsigned int num;
659 unsigned n;
660 __be64 *ptr;
661
662 gfs2_log_lock(sdp);
663 list_sort(NULL, blist, blocknr_cmp);
664 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
665 while(total) {
666 num = total;
667 if (total > limit)
668 num = limit;
669 gfs2_log_unlock(sdp);
670 page = gfs2_get_log_desc(sdp,
671 is_databuf ? GFS2_LOG_DESC_JDATA :
672 GFS2_LOG_DESC_METADATA, num + 1, num);
673 ld = page_address(page);
674 gfs2_log_lock(sdp);
675 ptr = (__be64 *)(ld + 1);
676
677 n = 0;
678 list_for_each_entry_continue(bd1, blist, bd_list) {
679 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
680 if (is_databuf) {
681 gfs2_check_magic(bd1->bd_bh);
682 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
683 }
684 if (++n >= num)
685 break;
686 }
687
688 gfs2_log_unlock(sdp);
689 gfs2_log_write_page(sdp, page);
690 gfs2_log_lock(sdp);
691
692 n = 0;
693 list_for_each_entry_continue(bd2, blist, bd_list) {
694 get_bh(bd2->bd_bh);
695 gfs2_log_unlock(sdp);
696 lock_buffer(bd2->bd_bh);
697
698 if (buffer_escaped(bd2->bd_bh)) {
699 void *kaddr;
700 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
701 ptr = page_address(page);
702 kaddr = kmap_atomic(bd2->bd_bh->b_page);
703 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
704 bd2->bd_bh->b_size);
705 kunmap_atomic(kaddr);
706 *(__be32 *)ptr = 0;
707 clear_buffer_escaped(bd2->bd_bh);
708 unlock_buffer(bd2->bd_bh);
709 brelse(bd2->bd_bh);
710 gfs2_log_write_page(sdp, page);
711 } else {
712 gfs2_log_write_bh(sdp, bd2->bd_bh);
713 }
714 gfs2_log_lock(sdp);
715 if (++n >= num)
716 break;
717 }
718
719 BUG_ON(total < num);
720 total -= num;
721 }
722 gfs2_log_unlock(sdp);
723}
724
725static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
726{
727 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
728 unsigned int nbuf;
729 if (tr == NULL)
730 return;
731 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
732 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
733}
734
735static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
736{
737 struct list_head *head;
738 struct gfs2_bufdata *bd;
739
740 if (tr == NULL)
741 return;
742
743 head = &tr->tr_buf;
744 while (!list_empty(head)) {
745 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
746 list_del_init(&bd->bd_list);
747 gfs2_unpin(sdp, bd->bd_bh, tr);
748 }
749}
750
751static void buf_lo_before_scan(struct gfs2_jdesc *jd,
752 struct gfs2_log_header_host *head, int pass)
753{
754 if (pass != 0)
755 return;
756
757 jd->jd_found_blocks = 0;
758 jd->jd_replayed_blocks = 0;
759}
760
761#define obsolete_rgrp_replay \
762"Replaying 0x%llx from jid=%d/0x%llx but we already have a bh!\n"
763#define obsolete_rgrp_replay2 \
764"busy:%d, pinned:%d rg_gen:0x%llx, j_gen:0x%llx\n"
765
766static void obsolete_rgrp(struct gfs2_jdesc *jd, struct buffer_head *bh_log,
767 u64 blkno)
768{
769 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
770 struct gfs2_rgrpd *rgd;
771 struct gfs2_rgrp *jrgd = (struct gfs2_rgrp *)bh_log->b_data;
772
773 rgd = gfs2_blk2rgrpd(sdp, blkno, false);
774 if (rgd && rgd->rd_addr == blkno &&
775 rgd->rd_bits && rgd->rd_bits->bi_bh) {
776 fs_info(sdp, obsolete_rgrp_replay, (unsigned long long)blkno,
777 jd->jd_jid, bh_log->b_blocknr);
778 fs_info(sdp, obsolete_rgrp_replay2,
779 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
780 buffer_pinned(rgd->rd_bits->bi_bh),
781 rgd->rd_igeneration,
782 be64_to_cpu(jrgd->rg_igeneration));
783 gfs2_dump_glock(NULL, rgd->rd_gl, true);
784 }
785}
786
787static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
788 struct gfs2_log_descriptor *ld, __be64 *ptr,
789 int pass)
790{
791 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
792 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
793 struct gfs2_glock *gl = ip->i_gl;
794 unsigned int blks = be32_to_cpu(ld->ld_data1);
795 struct buffer_head *bh_log, *bh_ip;
796 u64 blkno;
797 int error = 0;
798
799 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
800 return 0;
801
802 gfs2_replay_incr_blk(jd, &start);
803
804 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
805 blkno = be64_to_cpu(*ptr++);
806
807 jd->jd_found_blocks++;
808
809 if (gfs2_revoke_check(jd, blkno, start))
810 continue;
811
812 error = gfs2_replay_read_block(jd, start, &bh_log);
813 if (error)
814 return error;
815
816 bh_ip = gfs2_meta_new(gl, blkno);
817 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
818
819 if (gfs2_meta_check(sdp, bh_ip))
820 error = -EIO;
821 else {
822 struct gfs2_meta_header *mh =
823 (struct gfs2_meta_header *)bh_ip->b_data;
824
825 if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG))
826 obsolete_rgrp(jd, bh_log, blkno);
827
828 mark_buffer_dirty(bh_ip);
829 }
830 brelse(bh_log);
831 brelse(bh_ip);
832
833 if (error)
834 break;
835
836 jd->jd_replayed_blocks++;
837 }
838
839 return error;
840}
841
842static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
843{
844 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
845 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
846
847 if (error) {
848 gfs2_inode_metasync(ip->i_gl);
849 return;
850 }
851 if (pass != 1)
852 return;
853
854 gfs2_inode_metasync(ip->i_gl);
855
856 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
857 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
858}
859
860static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
861{
862 struct gfs2_meta_header *mh;
863 unsigned int offset;
864 struct list_head *head = &sdp->sd_log_revokes;
865 struct gfs2_bufdata *bd;
866 struct page *page;
867 unsigned int length;
868
869 gfs2_flush_revokes(sdp);
870 if (!sdp->sd_log_num_revoke)
871 return;
872
873 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
874 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
875 offset = sizeof(struct gfs2_log_descriptor);
876
877 list_for_each_entry(bd, head, bd_list) {
878 sdp->sd_log_num_revoke--;
879
880 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
881 gfs2_log_write_page(sdp, page);
882 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
883 mh = page_address(page);
884 clear_page(mh);
885 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
886 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
887 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
888 offset = sizeof(struct gfs2_meta_header);
889 }
890
891 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
892 offset += sizeof(u64);
893 }
894 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
895
896 gfs2_log_write_page(sdp, page);
897}
898
899void gfs2_drain_revokes(struct gfs2_sbd *sdp)
900{
901 struct list_head *head = &sdp->sd_log_revokes;
902 struct gfs2_bufdata *bd;
903 struct gfs2_glock *gl;
904
905 while (!list_empty(head)) {
906 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
907 list_del_init(&bd->bd_list);
908 gl = bd->bd_gl;
909 gfs2_glock_remove_revoke(gl);
910 kmem_cache_free(gfs2_bufdata_cachep, bd);
911 }
912}
913
914static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
915{
916 gfs2_drain_revokes(sdp);
917}
918
919static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
920 struct gfs2_log_header_host *head, int pass)
921{
922 if (pass != 0)
923 return;
924
925 jd->jd_found_revokes = 0;
926 jd->jd_replay_tail = head->lh_tail;
927}
928
929static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
930 struct gfs2_log_descriptor *ld, __be64 *ptr,
931 int pass)
932{
933 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
934 unsigned int blks = be32_to_cpu(ld->ld_length);
935 unsigned int revokes = be32_to_cpu(ld->ld_data1);
936 struct buffer_head *bh;
937 unsigned int offset;
938 u64 blkno;
939 int first = 1;
940 int error;
941
942 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
943 return 0;
944
945 offset = sizeof(struct gfs2_log_descriptor);
946
947 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
948 error = gfs2_replay_read_block(jd, start, &bh);
949 if (error)
950 return error;
951
952 if (!first)
953 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
954
955 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
956 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
957
958 error = gfs2_revoke_add(jd, blkno, start);
959 if (error < 0) {
960 brelse(bh);
961 return error;
962 }
963 else if (error)
964 jd->jd_found_revokes++;
965
966 if (!--revokes)
967 break;
968 offset += sizeof(u64);
969 }
970
971 brelse(bh);
972 offset = sizeof(struct gfs2_meta_header);
973 first = 0;
974 }
975
976 return 0;
977}
978
979static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
980{
981 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
982
983 if (error) {
984 gfs2_revoke_clean(jd);
985 return;
986 }
987 if (pass != 1)
988 return;
989
990 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
991 jd->jd_jid, jd->jd_found_revokes);
992
993 gfs2_revoke_clean(jd);
994}
995
996/**
997 * databuf_lo_before_commit - Scan the data buffers, writing as we go
998 * @sdp: The filesystem
999 * @tr: The system transaction being flushed
1000 */
1001
1002static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1003{
1004 unsigned int limit = databuf_limit(sdp);
1005 unsigned int nbuf;
1006 if (tr == NULL)
1007 return;
1008 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
1009 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
1010}
1011
1012static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
1013 struct gfs2_log_descriptor *ld,
1014 __be64 *ptr, int pass)
1015{
1016 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1017 struct gfs2_glock *gl = ip->i_gl;
1018 unsigned int blks = be32_to_cpu(ld->ld_data1);
1019 struct buffer_head *bh_log, *bh_ip;
1020 u64 blkno;
1021 u64 esc;
1022 int error = 0;
1023
1024 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1025 return 0;
1026
1027 gfs2_replay_incr_blk(jd, &start);
1028 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1029 blkno = be64_to_cpu(*ptr++);
1030 esc = be64_to_cpu(*ptr++);
1031
1032 jd->jd_found_blocks++;
1033
1034 if (gfs2_revoke_check(jd, blkno, start))
1035 continue;
1036
1037 error = gfs2_replay_read_block(jd, start, &bh_log);
1038 if (error)
1039 return error;
1040
1041 bh_ip = gfs2_meta_new(gl, blkno);
1042 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1043
1044 /* Unescape */
1045 if (esc) {
1046 __be32 *eptr = (__be32 *)bh_ip->b_data;
1047 *eptr = cpu_to_be32(GFS2_MAGIC);
1048 }
1049 mark_buffer_dirty(bh_ip);
1050
1051 brelse(bh_log);
1052 brelse(bh_ip);
1053
1054 jd->jd_replayed_blocks++;
1055 }
1056
1057 return error;
1058}
1059
1060/* FIXME: sort out accounting for log blocks etc. */
1061
1062static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1063{
1064 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1065 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1066
1067 if (error) {
1068 gfs2_inode_metasync(ip->i_gl);
1069 return;
1070 }
1071 if (pass != 1)
1072 return;
1073
1074 /* data sync? */
1075 gfs2_inode_metasync(ip->i_gl);
1076
1077 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1078 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1079}
1080
1081static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1082{
1083 struct list_head *head;
1084 struct gfs2_bufdata *bd;
1085
1086 if (tr == NULL)
1087 return;
1088
1089 head = &tr->tr_databuf;
1090 while (!list_empty(head)) {
1091 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1092 list_del_init(&bd->bd_list);
1093 gfs2_unpin(sdp, bd->bd_bh, tr);
1094 }
1095}
1096
1097
1098static const struct gfs2_log_operations gfs2_buf_lops = {
1099 .lo_before_commit = buf_lo_before_commit,
1100 .lo_after_commit = buf_lo_after_commit,
1101 .lo_before_scan = buf_lo_before_scan,
1102 .lo_scan_elements = buf_lo_scan_elements,
1103 .lo_after_scan = buf_lo_after_scan,
1104 .lo_name = "buf",
1105};
1106
1107static const struct gfs2_log_operations gfs2_revoke_lops = {
1108 .lo_before_commit = revoke_lo_before_commit,
1109 .lo_after_commit = revoke_lo_after_commit,
1110 .lo_before_scan = revoke_lo_before_scan,
1111 .lo_scan_elements = revoke_lo_scan_elements,
1112 .lo_after_scan = revoke_lo_after_scan,
1113 .lo_name = "revoke",
1114};
1115
1116static const struct gfs2_log_operations gfs2_databuf_lops = {
1117 .lo_before_commit = databuf_lo_before_commit,
1118 .lo_after_commit = databuf_lo_after_commit,
1119 .lo_scan_elements = databuf_lo_scan_elements,
1120 .lo_after_scan = databuf_lo_after_scan,
1121 .lo_name = "databuf",
1122};
1123
1124const struct gfs2_log_operations *gfs2_log_ops[] = {
1125 &gfs2_databuf_lops,
1126 &gfs2_buf_lops,
1127 &gfs2_revoke_lops,
1128 NULL,
1129};
1130